Compute Is Becoming a Strategic Asset. America’s Grid Wasn’t Built for It.

Every era of strategic competition has had a material foundation. In the twentieth century it was oil, steel and shipbuilding capacity. In the early twenty-first it was semiconductor fabrication. The emerging consensus in Washington’s national security community is that the next such foundation is already visible: the compute, energy and data centre infrastructure on which advanced artificial intelligence depends.

That consensus marks a quiet but consequential shift. The first phase of the AI policy debate was dominated by questions about the technology itself: model safety, misuse, bias, cyber risk. Those questions remain live. But security planners are increasingly focused on a more elementary problem. The United States may lead the world in AI models and chip design while lacking the physical capacity, above all electric power, to deploy that lead at scale.

Why the security establishment is paying attention

The national security stakes are not abstract. Advanced AI systems are expected to reshape intelligence analysis, cyber defence, military logistics, scientific research and the resilience of financial and critical infrastructure. None of those capabilities exists in software alone. Each depends on large, power-hungry facilities connected to a grid that was never designed for them, supplied through specialised industrial chains, and sited in communities whose consent cannot be assumed.

The strain is now measurable. PJM Interconnection, the country’s largest grid operator, has been managing record electricity demand driven in part by rapid data centre growth. In June, federal energy regulators ordered six regional grid operators to justify or reform how very large power users connect to the system, pushing for faster studies, clearer cost allocation and better planning around co-located generation. The Department of Energy now draws an explicit line between data centres, energy capacity, AI leadership and national security, and private capital is following the same logic into power platforms purpose-built for AI demand.

“Strategic assets are the things a country cannot improvise in a crisis,” said Samir “Sam” Tabar, chief executive of WhiteFiber and Bit Digital. “You cannot improvise a gigawatt of firm power, a substation, or a data centre campus. They take years to build. That is what makes compute capacity a national security question rather than just a commercial one.”

The bottleneck has moved

For most of the past decade, the binding constraint in AI was assumed to be chips or talent. Increasingly, it is speed to power. In several regions, developers can raise capital and secure GPUs faster than they can obtain a grid connection. Interconnection queues built for a handful of large projects a year are absorbing dozens at once. The result is a competition that looks less like a software race and more like an industrial one, fought over land, transmission capacity, cooling, fibre, permitting and execution.

WhiteFiber, a Nasdaq-listed AI infrastructure company, operates in the segment where these constraints bind first. Its model spans power procurement, data centre design, colocation and GPU cloud services, the layer of the stack that turns capital and chips into usable compute.

Recent activity at the company illustrates how capacity is actually being built. In May, WhiteFiber announced a $100 million delayed-draw credit facility to support growth across its data centre and cloud businesses, including a high-performance computing site in North Carolina.

It also signed a five-year, $160 million AI compute agreement supporting expansion into the Paris region, a signal that allied demand for Nvidia GPU-powered capacity is growing alongside American demand. The transactions are commercial, but the pattern behind them is strategic: long-duration customers, specialised financing and access to power are now the inputs that determine who can field AI capability at all.

“There is a habit of treating the cloud as if it were weightless,” Tabar said. “In reality, every serious AI deployment resolves to a physical site with a grid connection, a water and cooling plan, a supply chain and a security perimeter. If you care about where AI capability sits geopolitically, you have to care about where those sites are and how fast they can be built.”

Resilience, not just capacity

A security-minded view of the buildout raises questions that go beyond megawatts. Concentrating frontier compute in a small number of regions creates single points of failure.

Data centre campuses supporting defence, intelligence or critical-infrastructure workloads are attractive targets for cyber and physical disruption, and their dependencies, from transformers to specialised cooling equipment, run through global supply chains that are not uniformly secure. A resilient posture would treat siting diversity, hardened facilities, secure supply chains and cybersecurity standards as part of the same policy problem as generation and transmission.

It would also distinguish between categories of infrastructure. Routine cloud expansion and a frontier training campus with national security relevance are not the same asset and should not be governed as if they were.

The domestic bargain is a security issue too

There is a further constraint that planners ignore at their peril: public consent. If communities conclude that AI demand will raise household electricity bills, deplete water or strain local systems without local benefit, the buildout will slow regardless of its strategic importance.

The White House’s Ratepayer Protection Pledge speaks to that risk, calling on major AI and data centre companies to build, bring or buy the power their facilities need and to fund new delivery infrastructure rather than shifting costs onto ordinary customers. PJM has weighed proposals along similar lines, including requiring data centres to fund new supply or accept curtailment when the system is stressed.

Seen through a security lens, cost transparency and community benefit are not concessions that slow the mission. They are what makes a multi-decade buildout politically sustainable.

“Durability matters as much as speed,” Tabar said. “An infrastructure programme that alienates the communities hosting it, or that leaves households paying for private demand, will stall. The strategic requirement is capacity that can be built quickly and defended politically for twenty years.”

Where the contest will be decided

Models will continue to dominate headlines, and semiconductors will remain the most visible chokepoint in the supply chain. But the decisive terrain of the AI competition is shifting toward substations, transmission queues, energy contracts and data centre campuses, and toward the permitting, financing and execution capacity that determines how fast they materialise. Rival powers are not treating that layer as an afterthought; they are aligning energy, industrial and security policy around it.

The United States retains real advantages: the leading model developers, the deepest capital markets and a private infrastructure sector already mobilising. Whether those advantages translate into durable leadership will depend on a question that is no longer technological. It is whether America is prepared to treat compute, energy and data centre infrastructure as what they have become: strategic national assets.